AIR#32 - April 05, 2024
Good morning, AI enthusiasts! As you kickstart your day with a warm cup of coffee, today's edition of AIR: The AI Recon is here to serve you the hottest slices of AI news, seasoned with groundbreaking developments and a pinch of controversy. Leading the pack, we have Nvidia's AI-driven data center processors now accounting for a whopping 78% of its revenue, showcasing a seismic shift in the tech landscape where AI hardware demand is not just growing; it's exploding. This story is a testament to the AI boom's profound impact on industry giants, reshaping priorities and fueling an insatiable appetite for more powerful, efficient AI solutions.
But it's not all about the hardware. Today, we also dive into the ethical whirlpool stirred up by Israel's use of AI "Lavender" in generating assassination targets, a chilling reminder of AI's potential to not only innovate but also intimidate. This tale of "AI-assisted genocide" sparks outrage and raises pressing questions about the moral compass guiding AI's application in sensitive, life-altering scenarios. It's a stark contrast to the tech triumphs we often celebrate, urging us to ponder the delicate balance between advancement and ethics.
On a lighter note, we're also exploring the quirky side of AI with the advent of AI-generated lawyers issuing fake DMCA threats. It's a wild ride into the realm of AI-enabled scams that underscores the creative (and sometimes dubious) uses of AI in the digital age. As you savor today's stories, let them inspire, challenge, and perhaps even redefine your understanding of AI's vast landscape. From the boardrooms of Nvidia to the controversial fields of Gaza, and the bizarre world of AI scams, there's never a dull moment in the ever-evolving saga of artificial intelligence. Let's dive in!
Business
Cloudflare Unveils Serverless AI on Global Edge Network, Challenging AWS
Cloudflare shakes up the AI scene with serverless AI on its global network, directly challenging AWS with accessible, scalable AI solutions.
Google Launches AI-Generated Search Answers in UK Trial
Google trials AI-generated search answers in the UK, aiming to enhance user experience while maintaining site traffic and addressing potential biases and environmental concerns.
YouTube Accuses OpenAI of Potential Terms of Service Breach with Sora Training
YouTube warns OpenAI: using its videos to train Sora AI could violate terms of service.
PitchPal: Use GPT-4 to Craft Your YC Application Effortlessly
Dreaming of YC? PitchPal uses GPT-4 to tailor your startup application, making funding dreams a breeze. ✨🚀
Google May Charge for AI-Enhanced Internet Searches
Google mulls charging for AI-enhanced searches, potentially shifting to a subscription model to cover high AI costs.
AI Boom Propels Nvidia's Data Center Processors to 78% of Revenue
Nvidia's AI-driven data center processors now dominate its revenue, making up 78%, a massive leap from GPUs. Demand for AI hardware skyrockets, pushing $22.1B in sales.
Washington State Judge Bans AI-Enhanced Video Evidence in Landmark Ruling
Washington judge bans AI-enhanced video in court, citing confusion and reliability issues in a groundbreaking ruling.
Apple Shifts Focus to Home Robotics After Electric Car Project Stalls
Apple pivots to home robotics, exploring personal robots after electric car project stalls.
🔥 AI-Generated Lawyers Issue Fake DMCA Threats for SEO Scams
Beware: Fake AI law firms are issuing bogus DMCA threats for SEO scams, using AI-generated images and nonsensical legal demands.
Engineering
[GitHub] AMD's Riallto: Explore Ryzen AI NPU with Open-Source Framework
AMD launches Riallto, an open-source framework to explore Ryzen AI NPU, complete with Jupyter notebooks for beginners.
[Github] Dolphin 2.8 Mistral 7B v0.2: Uncensored LLM by Eric Hartford & Cognitive Computations
Dolphin 2.8, an uncensored LLM by Eric Hartford, boasts diverse skills & advises caution with its use. Funded by Crusoe Cloud & others, it's Apache 2.0 licensed.
Bharath Ramsundar on X: LLMs as Interpolative Vector Databases and Their Scientific Prospects
Bharath Ramsundar views LLMs as vast, interpolative vector databases with potential yet untapped scientific prospects, needing richer biases for breakthroughs.
The Fine Art of Human Prompt Engineering: Chatting with People Like AI
Discover the ancient art of human prompt engineering: mastering conversation techniques to engage and understand people like never before.
Awni Hannun Achieves 100 Tokens/Sec with 4-bit Mistral 7B on M2 Ultra, Surpassing llama.cpp
Awni Hannun sets a new speed record: 100 tokens/sec using 4-bit Mistral 7B on M2 Ultra, outpacing llama.cpp.
DataStax Buys Langflow to Boost Generative AI Development
DataStax acquires Langflow, boosting AI app development with easier building and rapid iteration, aiming to democratize AI creation.
🔥 [Paper] JetMoE-8B: Surpassing LLaMA2 with $0.1M Budget
JetMoE-8B AI model outshines LLaMA2 with just $0.1M, proving high performance doesn't need big budgets. Open-source and low compute needs make it accessible for all.
Academic
[Paper] Do Transformers Anticipate Future Tokens? Insights from Pre-caching and Breadcrumbs
Exploring if AI predicts future words, researchers find evidence for "pre-caching" and "breadcrumbs" strategies in language models.
[Paper] Mixture-of-Depths: Efficient Compute Allocation in Transformer Models
New study introduces Mixture-of-Depths, slashing AI model training time by 50% by smartly allocating compute resources.
Explosive Report: Israeli Army's AI "Lavender" Mass-Generates Assassination Targets in Gaza
Explosive report reveals Israeli AI "Lavender" mass-generates Palestinian assassination targets, leading to widespread civilian casualties.
Meta's AI Struggles to Generate Images of Interracial Couples
Meta's AI fails at generating images of interracial couples, showing bias and a narrow view of diversity.
'AI-assisted genocide': Israel's AI Database for Gaza Kill Lists Sparks Outrage
Israel's AI "Lavender" linked to Gaza kill lists, sparking war crime concerns; thousands dead as AI errors and policy fuel civilian casualties.
W3C's Guide on AI's Web Impact and Standardization Needs
W3C explores AI's web impact, highlighting the need for standards in consent, labeling AI content, data privacy, and interoperability to manage ethical, societal, and technical challenges.
🔥 [Paper] Think-and-Execute: Enhancing LMs' Algorithmic Reasoning with Pseudocode Simulation
Simulating pseudocode execution in LMs boosts their algorithmic reasoning, proving task-level logic and tailored code are key to problem-solving.